Soa: Modify Infer Locations

.devcontainer/Dockerfile

@@ -31,8 +31,8 @@ RUN echo "export PATH=${PATH}" >> /home/${USERNAME}/.profile
 # install ghcup
 RUN curl --proto '=https' --tlsv1.2 -sSf https://get-ghcup.haskell.org | sh
 
-ARG GHC=9.4.6
-ARG CABAL=3.8.1.0
+ARG GHC=9.10.1
+ARG CABAL=3.12.1.0
 ARG STACK=2.9.3
 ARG HLS=recommended
 

gibbon-compiler/src/Gibbon/Common.hs

@@ -8,7 +8,7 @@
 module Gibbon.Common
        (
          -- * Variables
-         Var(..), LocVar(..), Location, FieldIndex
+         Var(..), LocVar(..), Location, FieldIndex, DataCon
        , RegVar, fromVar, toVar, varAppend, toEndV, toEndVLoc, toSeqV, cleanFunName
        , TyVar(..), isUserTv
        , Symbol, intern, unintern
@@ -66,7 +66,6 @@ import System.IO.Unsafe ( unsafePerformIO )
 import System.Random    ( randomIO )
 import Debug.Trace
 import Language.C.Quote.CUDA (ToIdent, toIdent)
-
 import Gibbon.DynFlags
 
 --------------------------------------------------------------------------------
@@ -138,12 +137,19 @@ toSeqV v = varAppend v (toVar "_seq")
 
 -- | A location variable stores the abstract location. 
 type Location = Var
-
--- | The position or index of a field in a data constructor value. 
-type FieldIndex = Int
-
-data LocVar = Single Location
-  deriving (Show, Ord, Eq, Read, Generic, NFData, Out)
+-- | The position or index of a field in a data constructor value.
+type FieldIndex = Int 
+-- | A data constructor is a String type in the compiler
+type DataCon = String
+
+-- | Single: For storing a single location, useful for adding a cursor in a region. 
+-- | SoA: A location signature for a structure of arrays representation. 
+--        The first location points to a location in the data constructor buffer. 
+--        The list includes locations for each field and a tuple storing 
+--        information about which data constructor and corresponding index the field 
+--        comes from. 
+data LocVar = Single Location | SoA Location [((DataCon, FieldIndex), Location)]
+                deriving (Show, Ord, Eq, Read, Generic, NFData, Out)
 
 -- | Abstract region variables.
 type RegVar = Var

gibbon-compiler/src/Gibbon/L2/Syntax.hs

@@ -185,16 +185,23 @@ data E2Ext loc dec
 
 -- | Define a location in terms of a different location.
 data PreLocExp loc = StartOfRegionLE Region
-                   | AfterConstantLE Int  -- Number of bytes after.
-                                     loc  -- Location which this location is offset from.
-                   | AfterVariableLE Var  -- Name of variable v. This loc is size(v) bytes after.
-                                     loc  -- Location which this location is offset from.
-                                     Bool -- Whether it's running in a stolen continuation i.e
-                                          -- whether this should return an index in a fresh region or not.
-                                          -- It's True by default and flipped by ParAlloc if required.
+                   | AfterConstantLE  Int  -- Number of bytes after. (In case of an SoA loc, this is the offset into the data constructor buffer)
+                                      loc  -- Location which this location is offset from.
+
+                   | AfterVariableLE  Var  -- Name of variable v. This loc is size(v) bytes after.
+                                      loc  -- Location which this location is offset from.
+                                      Bool -- Whether it's running in a stolen continuation i.e
+                                           -- whether this should return an index in a fresh region or not.
+                                           -- It's True by default and flipped by ParAlloc if required.
                    | InRegionLE Region
                    | FreeLE
                    | FromEndLE  loc
+
+                   | AfterSoALE (PreLocExp loc) [PreLocExp loc] loc  -- Compute new SoA location from an old SoA location 
+								     -- (PreLocExp loc) -> expression for arithmetic on data constructor buffer 
+								     -- [PreLocExp loc] -> expressions for arithmetic on each field location 
+								     -- loc, store the old loc, why? -- capture more metadata, also style
+
   deriving (Read, Show, Eq, Ord, Functor, Generic, NFData)
 
 type LocExp = PreLocExp LocVar
@@ -231,7 +238,7 @@ instance FreeVars LocExp where
   gFreeVars e =
     case e of
       AfterConstantLE _ loc   -> S.singleton $ unwrapLocVar loc
-      AfterVariableLE v loc _ -> S.fromList $ [v, unwrapLocVar loc]
+      AfterVariableLE v loc _ -> S.fromList [v, unwrapLocVar loc]
       _ -> S.empty
 
 instance (Out l, Out d, Show l, Show d) => Expression (E2Ext l d) where
@@ -479,6 +486,17 @@ data Region = GlobR Var Multiplicity -- ^ A global region with lifetime equal to
                                      --   are no free locations in the program.
   deriving (Read,Show,Eq,Ord, Generic)
 
+{- 
+   Why a new datatype? -- Well, For a location, we changed the exisiting datatype, i.e. LocVar.
+   That's acceptable, because locations are just cursors into regions, they are an indices into 
+   regions which don't necessarily signify an entity/allocated object. 
+
+   A region encapsulates an allocated space and i'd like a region to represent a unit of space. 
+   We can create more "complex" regions using that unit, for instance a SoA region. 
+   In additon, it preserves statements like "is region A within region B?"
+   It would be tough to formalize what is means by saying an SoA region exists within another region 
+   (VarR x). Since an SoA region contains multiple regions.
+-}
 data ExtendedRegion = AoSR Region                                   -- ^ A simple "flat" region where the datatype 
                                                                     --   will reside in an array of structure representation.
                     | SoAR Region [((DataCon, FieldIndex), Region)] -- ^ A complex region representation for a datatype 

gibbon-compiler/src/Gibbon/L2/Typecheck.hs

@@ -780,6 +780,7 @@ tcExp ddfs env funs constrs regs tstatein exp =
                        (ty,tstate2) <- tcExp ddfs env' funs constrs1 regs tstate1 e
                        tstate3 <- removeLoc exp tstate2 (Single loc)
                        return (ty,tstate3)
+                {-TODO handle what needs to happen with the wildcard argument, list of offsets in case of soa -}
                 AfterConstantLE i l1 ->
                      do r <- getRegion exp constrs l1
                         let tstate1 = extendTS (Single loc) (Output,True) $ setAfter l1 tstatein

gibbon-compiler/src/Gibbon/L4/Syntax.hs

@@ -103,7 +103,7 @@ type Label = Var
 type SymTable = M.Map Word16 String
 
 type InfoTable = (M.Map L.TyCon TyConInfo)
-type TyConInfo = M.Map L.DataCon DataConInfo
+type TyConInfo = M.Map DataCon DataConInfo
 
 data DataConInfo = DataConInfo
   { dcon_tag :: Tag

gibbon-compiler/src/Gibbon/Language/Syntax.hs

@@ -14,7 +14,7 @@
 module Gibbon.Language.Syntax
   (
     -- * Datatype definitions
-    DDefs, DataCon, TyCon, Tag, IsBoxed, DDef(..)
+    DDefs, TyCon, Tag, IsBoxed, DDef(..)
   , lookupDDef, getConOrdering, getTyOfDataCon, lookupDataCon, lkp
   , lookupDataCon', insertDD, emptyDD, fromListDD, isVoidDDef
 
@@ -78,7 +78,6 @@ import           Gibbon.Common
 
 type DDefs a = M.Map Var (DDef a)
 
-type DataCon = String
 type TyCon   = String
 type Tag     = Word8
 

gibbon-compiler/src/Gibbon/NewL2/Syntax.hs

@@ -140,7 +140,7 @@ instance FreeVars LocExp where
   gFreeVars e =
     case e of
       Old.AfterConstantLE _ loc  -> S.singleton $ unwrapLocVar (toLocVar loc)
-      Old.AfterVariableLE v loc _ -> S.fromList $ [v, unwrapLocVar (toLocVar loc)] 
+      Old.AfterVariableLE v loc _ -> S.fromList [v, unwrapLocVar (toLocVar loc)] 
       _ -> S.empty
 
 

gibbon-compiler/src/Gibbon/Passes/CalculateBounds.hs

@@ -180,9 +180,11 @@ calculateBoundsExp ddefs env2 varSzEnv varLocEnv locRegEnv locOffEnv regSzEnv re
                   else do
                     let (re, off) = case locExp of
                           (StartOfRegionLE r          ) -> (regionToVar r, BoundedSize 0)
+                          -- [2024.12.04] VS: currently discarding offsets for SoA representation
                           (AfterConstantLE n l  ) -> (locRegEnv # l, locOffEnv # l <> BoundedSize n)
                           -- [2022.12.26] CSK: the lookup in varSzEnv always fails since the
                           -- pass never inserts anything into it. Disabling it for now.
+                          -- [2024.12.04] VS: currently discarding offsets for SoA representation
                           (AfterVariableLE v l _) -> (locRegEnv # l, locOffEnv # (varLocEnv # v)) -- <> varSzEnv # v
                           (InRegionLE r         ) -> (regionToVar r, Undefined)
                           (FromEndLE  l         ) -> (locRegEnv # l, Undefined)

gibbon-compiler/src/Gibbon/Passes/Cursorize.hs

@@ -699,6 +699,7 @@ cursorizeLocExp denv tenv senv lvar locExp =
       in if isBound ((toLocVar) loc) tenv
          then Right (rhs, [], tenv, senv)
          else Left$ M.insertWith (++) ((toLocVar) loc) [((unwrapLocVar lvar),[],CursorTy,rhs)] denv
+
     -- TODO: handle product types here
 
 {- [2018.03.07]: